저작자표시

(1)

저작자표시-비영리-변경금지 2.0 대한민국 이용자는 아래의 조건을 따르는 경우에 한하여 자유롭게

l 이 저작물을 복제, 배포, 전송, 전시, 공연 및 방송할 수 있습니다. 다음과 같은 조건을 따라야 합니다:

l 귀하는, 이 저작물의 재이용이나 배포의 경우, 이 저작물에 적용된 이용허락조건 을 명확하게 나타내어야 합니다.

저작권법에 따른 이용자의 권리는 위의 내용에 의하여 영향을 받지 않습니다. 이것은 이용허락규약(Legal Code)을 이해하기 쉽게 요약한 것입니다.

Disclaimer

저작자표시. 귀하는 원저작자를 표시하여야 합니다.

비영리. 귀하는 이 저작물을 영리 목적으로 이용할 수 없습니다.

변경금지. 귀하는 이 저작물을 개작, 변형 또는 가공할 수 없습니다.

(2)

Master’s Thesis of Economics

Demand for three major protein sources in Korea: Focusing on

terrestrial, seafood, and plant protein

주요 단백질원에 따른 식품 수요와 시장 성공 요인

February 2022

Graduate School of

Agricultural Economics and Rural Development Seoul National University

Regional Information Studies Major

Yeowoon Park

(3)

Demand for three major protein sources in Korea: Focusing on

terrestrial, seafood, and plant protein

Examiner Junghoon Moon

Submitting a master ’ s thesis of Economics February 2022

Graduate School of

Agricultural Economics and Rural Development Seoul National University

Regional Information Studies Major Yeowoon Park

Confirming the master ’ s thesis written by Yeowoon Park

February 2022

Chair (Seal)

Vice Chair (Seal)

Examiner (Seal)

(4)

i

Abstract

Despite recent dramatic growth in seafood consumption worldwide, knowledge about seafood protein demand is limited relative to other protein sources. The present study models demand across three major protein categories (i.e., terrestrial protein, seafood protein, and plant protein) in South Korea to fill this research gap. In the first essay, household panel data is used to model a quadratic almost ideal demand system (QUAIDS). The study takes this one step further by modeling demand within four seafood protein categories (i.e., fish, cephalopods, shellfish, and crustaceans).

Sociodemographic variables, including health-related factors, are incorporated in the demand models. The results indicate that seafood protein is in a complementary relationship with terrestrial protein while substituting for plant protein. All four seafood categories are a substitute for each other. Individuals who take their health seriously are likely to consume more seafood, particularly fish. In the second essay, the study investigates the factors influencing the success of protein-based ready meals in online grocery retail. Furthermore, it explores how the relationship between these factors and ready meal dollar sales performance varies based on the protein source of the ready meal. As a result of the study, it is found that the sales of protein-based ready meals are significantly higher when the product is stored at a freezing temperature, is a restaurant collaborated product, or is a private label product. In addition, in the case of protein-based ready meals, the results indicate that the higher the concentration of the category to which the product belongs, the higher the sales performance.

Keywords: Terrestrial protein, seafood protein, plant protein, QUAIDS, random effect panel regression, ready meal

Student Number: 2020-29889

(5)

ii

Preface

A move away from animal protein toward plant protein is becoming a global agenda. Food provision has a significant impact on the natural systems on which socioeconomic development depends, an effect that can be primarily attributed to the amount of protein consumed as well as the proportion of different protein sources consumed (de Boer & Aiking, 2019). Thus, research is needed to understand the demand for major protein sources.

While the literature on sustainable diets strongly supports a switch from animal protein consumption toward plant protein consumption, it is much less explicit about the role of seafood protein consumption in diets (Irz et al., 2018). Moreover, in most previous research on consumer protein demand, seafood protein has been either excluded or combined with all terrestrial protein (e.g., beef, pork, and chicken). This inappropriate aggregation of different types of animal protein can lead to biases in the estimations of price elasticities and associated specification problems concerning the identification of substitutes (Salvanes & Devoretz, 1997). To fill the above research gaps, the present study focuses on South Korean households’ demand for protein sources. In the first essay, the study models the consumer demand for three major pre-processed proteins. Next, it moves on to investigate the demand for processed protein, namely protein-based ready meal products. The sales performance of these protein-based ready meals is assessed from the retailer’s perspective.

In the first essay, the study estimates two demand models. First, the demand for three major protein sources (terrestrial, seafood, and plant protein) are estimated. Next, a model of the demand for seafood protein is estimated by disaggregating seafood into the broad taxonomic groups of fish, cephalopods, shellfish, and crustaceans. In addition, the study identifies sociodemographic and health-related

(6)

iii

factors affecting protein demand. The results show that terrestrial protein and seafood protein are complementary, while plant protein can be seen as a substitute for both animal proteins.

In the second essay, the sales performance of ready meals on an online grocery channel is modelled according to the protein source through random effect panel regression analyses. As with the first essay, the first model considers the entire ready meal samples (including all protein sources) to estimate a panel regression model and determine which protein-based product sales performance is relatively better. Further, this study divides the dataset according to the protein source (i.e., terrestrial protein, seafood protein, and plant protein) to identify the factors that influence the success of each type of ready meal.

The results of this study will provide marketers and retailers with practical implications regarding the type of value proposition that can raise product sales in modern society, where the importance of sustainable protein consumption is emphasized. Furthermore, this study intends to provide implications for policymakers regarding the drivers of a demand shift from animal protein to plant protein. From an academic perspective, several previous studies have been conducted on the relationship between animal protein and plant protein. Meanwhile, the current study goes beyond the dichotomous division of protein and further subdivides the animal protein sources typically perceived by consumers as heterogeneous. From a practical perspective, studying the relative status of processed protein compared to pre-processed protein in the market and further researching the success factors of processed protein products according to their source suggests significant implications for new protein-based product development and product assortment strategy.

(7)

iv

I. Essay 1: Modelling the demand across three major protein sources: focusing on

seafood protein

Chapter 1. Introduction

The worldwide consumption of seafood had been slowly increasing until the early 2010s¹. However, surprisingly, the consumption increased by 21% from 2015 to 2016, where per capita consumption, in particular, grew faster in Asia and Oceania compared to the other continents (OECD-FAO, 2016). Such rapid growth is driven by two factors: a change in supply and/or demand (Kidane &

Brækkan, 2021)².

Demand growth plays a critical role in global seafood consumption. Even if there is no productivity growth, the demand growth can raise the market price, which, in turn, quantity supplied and consumed to increase (Kidane & Brækkan, 2021). Therefore, research on demand growth is essential, with previous studies having highlighted this importance (e.g., Asche et al., 2011; Brækkan et al., 2018; Kidane & Brækkan, 2021). However, due to the methodological complexity required, the demand side of seafood research has not received as much attention as its supply side.

1 The average annual per capita consumption of seafood has increased by roughly 10kg, from 9.9kg in the 1960s to 19.2kg in 2012 (FAO, 2014)

2 On the supply side, the expansion of aquaculture production and technological innovation have led to an increase in seafood supply (Asche, 2008; Garlock et al., 2020; Kobayashi et al., 2015; Tveterås et al., 2012).

Aquaculture production is increasing rapidly, representing 52% of the seafood supply for human consumption in 2018 (Garlock et al., 2020).

Furthermore, innovations in alternative feed and selective breeding have led to production cost-savings and the decline of the market place (Asche, 2008; Guttormsen, 2002).

(11)

2

A multitude of factors complicate seafood demand estimation. In particular, consumer heterogeneity in relation to health is a key factor influencing consumer demand for seafood (Kaabia et al., 2001;

Mintert et al., 2001; Torrissen & Onozaka, 2017). It is also important to understand seafood demand relative to the substitute protein sources. However, knowledge regarding the role of seafood demand in the demand literature remains rather limited (Irz et al., 2018). In the previous studies on consumer demand for meat, seafood has been either excluded or combined with all other types of meat (e.g., beef, pork, and chicken). Even when seafood protein has been included as a research subject and compared with terrestrial protein, researchers have failed to compare demand between animal and plant-based protein. The inappropriate aggregation of different types of protein can lead to biases in the estimations of price elasticities and associated specification problems when attempting to identify substitutes (Salvanes & Devoretz, 1997).

To fill the above research gaps, the present study focuses on South Korean households’ demand for protein sources in terms of the following three main objectives: (1) to examine the demand interrelationships among terrestrial meat, seafood, and plant protein;

(2) to investigate the demand interrelationships within specific seafood protein categories; and (3) to identify sociodemographic factors, including health-related factors, that affect protein consumption.

To achieve these research objectives, this study estimates two demand systems. In the first demand system (M1 henceforth), the demand for three major protein sources (terrestrial, seafood, and plant protein) is modelled, while, in the second (M2), seafood is disaggregated into the broad taxonomic groups of fish, cephalopods, shellfish, and crustaceans. Terrestrial protein includes fresh and frozen red meat (beef and pork) and white meat (chicken) in various cuts. Seafood protein consists of all fresh and frozen fish, cephalopods, shellfish, and crustaceans. Plant protein includes tofu

(12)

3

and legumes. Our rationale for analysing the demand relationships among these categories is as follows: (1) they are the major protein categories consumed by South Korean households; (2) there is a scarcity in the literature regarding consumer preferences for them;

(3) it is plausible that the factors driving consumer demand for each respective category differ; and (4) hence, each category can be expected to provide different types of utility. Finally, it should be noted that the subject of the analyses is limited to uncooked protein.

This study conducts its empirical analyses using unique household panel data collected by South Korea’s Rural Development Administration (RDA). The dataset includes information on grocery purchases by consumer panels, and the current study uses the dataset to analyse protein consumption at the household level. This study specifies the demand systems for major protein sources by applying the quadratic almost ideal demand system (QUAIDS) model proposed by Banks et al. (1997). As QUAIDS can also incorporate sociodemographic demand shifters, this study specifically examines the effect of the presence of children, age, body mass index (BMI), and degree of health concern on expenditure shares of each protein sources. The results have both economic and marketing implications for which strategies could be adopted to influence seafood consumption.

The remainder of this paper is as follows. The next section explains the pertinent literature on seafood demand. In the following section, QUAIDS framework is described. The subsequent section introduces the data and empirical specifications of the estimated demand model. The paper then presents and discusses the empirical results. In the final section, a summary is provided along with some concluding remarks.

(13)

4

Chapter 2. Previous literature

Several previous studies have addressed the demand for seafood (e.g. Bronnmann, 2016; Bronnmann et al., 2016; Buason & Agnarsson, 2020; Huang, 2015; Schrobback et al., 2019; Surathkal et al., 2017;

Tabarestani et al., 2017; Toufique et al., 2018). The meta-analysis by Gallet (2009) showed that the majority of demand specifications are based on the almost ideal system (AIDS) of Deaton and Muellbauer (1980). In addition to these AIDS specifications, previous studies have estimated the demand for seafood using functional forms such as double-log, semi-log, Rotterdam, Translog, and S-Branch (Gallet, 2009).

Focusing on recent studies that used AIDS models, Bronnmann (2016) addressed the demand for wild and aquaculture whitefish on the German market by using a general form of AIDS with linear approximation (LA-AIDS). The results indicated that the demand for aquaculture whitefish is relatively elastic and that pangasius is a substitute for wild-caught species, namely cod, pollock, and Alaska pollock. Surathkal et al. (2017) modelled frozen seafood in the United States using LA-AIDS to estimate the demand relationships among three aggregate frozen seafood categories (breaded, entrees, and unbreaded), as well as these relationships when disaggregated as finfish and shellfish. The results revealed that finfish and shellfish are mutual substitutes. Buason and Agnarsson (2020) examined French household demand for fresh salmon, frozen Salmonidae, fresh cod, frozen whitefish, and other seafood by estimating a two-regime infrequency of purchase model (IPM) and frequency-adjusted AIDS model. The results showed significant relationships between sociodemographic factors (e.g., family size, age, region, and BMI) and demand for different seafood categories.

Concerning previous studies conducted using QUAIDS model, Bronnmann et al. (2016) used a two-step procedure to study

(14)

5

German household demand for six frozen seafood products: farmed and wild salmon, farmed and wild shrimp, redfish, and cultured pangasius. The results indicated that German seafood consumers are generally price-sensitive to salmon and shrimp, implying that the German seafood industry’s revenue could rise if the supply increases. Toufique et al. (2018) estimated demand for fish categorized by their origin (inland capture, marine capture, and aquaculture) in Bangladesh by employing the QUAIDS model. The results indicated that the income elasticities for fish from all sources are positive and that the demand for fish from all sources becomes elastic depending on the household income level. The implications of these results are in line with the results from the study by Bronnmann et al. (2016). Increasing supply at aquaculture and inland capture fisheries is crucial for food security.

This study presents a novel perspective of seafood demand that differs from the aforementioned literature. Given the previous literature on seafood demand, this study proposes using QUAIDS model to specify the demand systems of three major protein sources and four seafood protein categories to derive the expenditure, own- price, and cross-price elasticities. Overall, the subject of our analyses as well as the protein classification can be considered more representative of current protein consumption contexts.

Chapter 3. The QUAIDS framework

The QUAIDS model is an extension of the AIDS model developed by Deaton and Muellbauer (1980). Following criticisms of the AIDS approach for yielding biased and inconsistent estimates (Asche &

Wessells, 1997), Banks et al. (1997) improved the AIDS model by adding a quadratic expenditure term, resulting in the QUAIDS model.

QUAIDS is an acknowledged model in fishery demand studies (e.g., Bronnmann, 2016; Bronnmann et al., 2016; Dey et al., 2011; Toufique

(15)

6

et al., 2018) and is calculated as follows (for notational simplicity, this study omits the subscript for household and time):

𝑤_𝑖 = 𝛼_𝑖0+ ∑^𝑛_𝑗=1𝛾_𝑖𝑗ln(𝑝_𝑗) + 𝛽𝑖ln { ^𝐸

𝑎(𝑝)} +_𝑏(𝑝)^𝜃^𝑖 𝑙𝑛 { ^𝐸

𝑎(𝑝)}²+ 𝜀_𝑖. (1) Here, 𝑤_𝑖 denotes the expenditure share of the 𝑖^th protein category, resulting in three expenditure share equations for M1 and four expenditure share equations for M2. 𝐸 indicates the total expenditure of the protein categories in each demand model (M1 and M2). 𝛼_𝑖0, 𝛾_𝑖𝑗, 𝛽_𝑖, and 𝜃_𝑖 are the parameters to be estimated. 𝛽_𝑖 measures the linear income effect, 𝛾_𝑖𝑗 measures the non-linear income effect, and 𝜀𝑖 is an error term. 𝑙𝑛(𝑝𝑖) is calculated as the logarithm of the price. 𝑎(𝑝) and 𝑏(𝑝) are non-linear price aggregators. 𝑙𝑛𝑎(𝑝) has a transcendental logarithm form, and 𝑏(𝑝) is a Cobb-Douglas price aggregator:

𝑙𝑛𝑎(𝑝) = 𝛼₀+ ∑^𝑛_𝑖=1𝛼_𝑖𝑙𝑛(𝑝_𝑖) +¹

2∑^𝑛_𝑖=1∑^𝑛_𝑗=1𝑙𝑛(𝑝_𝑖)𝑙𝑛(𝑝_𝑗) (2) 𝑏(𝑝) = ∏^𝑛_𝑖=1𝑝_𝑖^(𝜃^𝑖⁾. (3)

The present study imposes three sets of theoretical restrictions:

adding up, homogeneity, and symmetry. Adding up requires that the budget shares sum up to unity, implying the following:

∑^𝑛_𝑖=1𝛼_𝑖 = 1, ∑^𝑛_𝑖=1𝛽_𝑖 = 0, ∑^𝑛_𝑖=1𝛾_𝑖𝑗 = 0. (4)

The homogeneity of degree zero in prices and total expenditure assures that if all prices and income are multiplied by a positive constant, the quantity demanded must remain unchanged and requires:

∑^𝑛_𝑗=1𝛾_𝑖𝑗 = 0, ∑^𝑛_𝑗=1𝜃_𝑖= 0, ∀𝑖. (5)

(16)

7

Slutsky symmetry deals with the substitution effect between goods and restricts the matrix of substitution effects to be symmetric. To illustrate, the coefficient of the price of the 𝑖^𝑡ℎgood (𝑙𝑛(𝑝_𝑖)) has the same value in the budget share equation of the 𝑗^𝑡ℎ good as the coefficient of 𝑙𝑛(𝑝_𝑗) (Deaton & Muellbauer, 1980). The symmetry restriction is satisfied by:

𝛾_𝑖𝑗 = 𝛾_𝑗𝑖, ∀𝑖 ≠ 𝑗. (6)

The present study derives expenditure elasticities, 𝑒_𝑖 , uncompensated price elasticities, 𝑒𝑢𝑖𝑗 , and compensated price elasticities, 𝑒𝑐_𝑖𝑗 , following Banks et al. (1997). Differentiating Equation (1) with respect to 𝑙𝑛𝐸 and 𝑙𝑛(𝑝𝑗), respectively, yields:

𝜇_𝑖 ≡ ^𝜕𝑤^𝑖

𝜕ln𝑚= 𝛽_𝑖+ ^2𝜃^𝑖

𝑏(𝑝)[𝑙𝑛 {_𝑎(𝑝)^𝐸 }] (7)

𝜇_𝑖𝑗≡ ^𝜕𝑤^𝑖

𝜕ln𝑝_𝑗= 𝛾_𝑖𝑗− 𝜇_𝑖(𝛼_𝑗+ ∑ 𝛾_𝑘 _𝑗𝑘𝑙𝑛𝑝_𝑘) −^𝜆^𝑖^𝛽^𝑗

𝑏(𝑝)[𝑙𝑛 { ^𝐸

𝑎(𝑝)}]². (8)

The expenditure elasticities, 𝑒_𝑖, are given by³ 𝑒_𝑖 = ^𝜇^𝑖

𝑤_𝑖+ 1. (10)

The uncompensated (Marshallian) price elasticity, 𝑒𝑢_𝑖𝑗, takes the income effect and substitution effect into account and is calculated by:

3 It must be noted that the expenditure elasticity is based on expenditure on protein sources as in the QUAIDS model and does not directly reflect the consumer responses to total expenditures.

(17)

8 𝑒𝑢_𝑖𝑗 =^𝜇^𝑖𝑗

𝑤_𝑖− 𝛿_𝑖𝑗 where 𝛿_𝑖𝑗 = {1 if 𝑖 = 𝑗

0 if 𝑖 ≠ 𝑗. (11)

The compensated (Hicksian) price elasticity 𝑒𝑐𝑖𝑗, capturing the pure substitution effect of the price change, is computed by:

𝑒𝑐_𝑖𝑗= 𝑒𝑢_𝑖𝑗+ 𝑒_𝑖𝑤_𝑗. (12)

Elasticities are a function of the parameter estimates of the demand system and expenditure shares. This study uses the sample means of the shares to calculate the elasticities.

Chapter 4. Data and empirical specification

In the empirical analyses, the current study used household panel data collected by the RDA. The dataset was collected from a randomly selected consumer panel using a stratified sampling method.

The respondents attached all their daily food purchase receipts to the housekeeping book and recorded the information, including purchase frequency from January 2015 to November 2019.

In addition to the purchase information, the dataset incorporates multiple sociodemographic variables. The choice of variables was determined following the previous literature (e.g., Buason &

Agnarsson, 2020; Dey et al., 2011) but was also dictated by the availability of variables in the data. The dataset includes a dummy variable for the presence of children, an ordinal variable for the age of the household head, and a logarithm of household income. To account for the health-related factors of the respondents, this study merged the dataset with supplemental questionnaire data on the respondents’ health concerns (HCs) and BMIs, which were distributed in 2015.

(18)

9

The following procedures describe how the merged dataset was prepared for econometric analysis. First, household panel data from a total of 835 households were collected. During the process, unusable observations containing missing values (two households) were removed from the dataset. Respondents who did not respond to the supplemental questionnaire were removed from the sample, yielding a total of 634 household samples for M1. For M2, three households without seafood purchase history from January 2015 to November 2019 were removed from the sample, yielding a total of 631 household samples. In summary, the final dataset consisted of 634 households with 35,654 observations for M1 and 631 households with 22,643 observations for M2.

Expenditures were aggregated into monthly figures for further matching with the price data. Since there was no price variable in the dataset, the price variable was created using expenditure shares and the Consumer Price Index (CPI). The final model included sociodemographic and seasonality variables, with endogeneity handled by utilizing a logarithm of income. In the following sections, the data management process and QUAIDS model specification are given in detail.

4.1. Demand system estimation in the absence of price data

Previous studies have used several approaches to compensate for a lack of price data (Castellón et al., 2015). Some consumer expenditure surveys have collected data on purchase quantities and expenditures, allowing for the calculation of unit values (i.e., expenditure divided by quantities), which are used as proxies for prices (e.g., Cox & Wohlgenant, 1986; Deaton, 1988). Another common approach has been incorporating external sources of price variability, such as the CPI, to account for missing prices (e.g., Bronnmann et al., 2016; Kastens & Brester, 1996; Kim et al., 2019;

(19)

10

Seale Jr. et al., 2003). However, studies conducted by Slesnick (2005) and Hoderlein and Mihaleva (2008) have found this approach problematic, as it does not account for household variability.

Considering this criticism expressed in previous literature, in this study, the price variables were generated using both the CPI and unit values.

First, the CPI⁴ was used to account for the price fluctuations of each category. The study deflated the unit values for January 2015 by the monthly CPI (base year =2015) as of January 2015 to yield the final price data. To match the category of protein sources provided by the CPI, first, the expenditures on proteins were grouped into 16 subgroups: domestic beef, imported beef, pork, chicken, mackerel, hairtail, croaker, pollock, squid, small octopus, abalone, oyster, clam, crab, tofu, and legumes. Despite the unavailability of shrimp’s CPI, the category was added to the analyses⁵ since it occupies a significant portion of South Korean seafood consumption, This resulted in a total of 17 subgroups.

This study generated its unit values by dividing the monthly expenditure by the monthly quantities of each good purchased on each shopping trip. This approach has been used in other demand studies, including the studies of Allais et al. (2010) and Bertail and Caillavet (2008). This approach not only accounts for household variability but also for cross-product differences. Before generating the unit values, all quantity observations in the upper 5% and lower 5% were dropped, since the data on the purchased quantity included a few outliers, which were probably due to errors in data recording.

Then, the unit values of January 2015 were deflated by the monthly

4 The CPI is managed by South Korea and is calculated monthly by an actual price survey of items. The categories of interest satisfy the following;

first, the average household spending per capita of national households is greater than a certain percentage; second, items represent the price of the same species group; and third, items are continuously priced in the market.

5 The CPI of shrimp is included in the survey from 2020.

(20)

11

CPI (base year =2015) as of January 2015 to yield the final price data.

Table 1 presents the descriptive statistics of the unit prices of each protein subgroup averaged over the years from January 2015 to November 2019. It can be seem that domestic beef has the highest unit price at about US$37/kg among subgroups of terrestrial meat, followed by chicken at about US$7/kg. Among the categories of seafood protein, abalone exhibits the highest unit price at about US$47/kg, followed by shrimp at about US$18/kg. Croaker records the lowest unit price among the categories of seafood protein; the average unit price for this species is about US$7/kg. The unit price of tofu is US$4/kg, which is the lowest price among all the protein subgroups.

(21)

12

Table 1. Descriptive statistics of unit prices.

Unit Price (US$/kg)^a Mean Std.Dev. Terrestrial protein

Beef

Domestic beef 37.055 2.400

Imported beef 19.856 0.720

Pork 12.690 0.782

Chicken 6.505 0.286

Seafood protein

Fish

Mackerel 12.680 0.636

Hairtail 10.649 0.908

Croaker 7.494 0.260

Pollock 6.700 0.135

Cephalopods

Squid 10.062 2.773

Small octopus 13.954 2.100

Shellfish

Abalone 47.483 4.236

Oyster 12.426 1.048

Clam 7.193 0.393

Crustaceans

Crab 12.772 1.462

Shrimp 18.471 2.170

Plant protein

Tofu 4.080 0.049

Legumes 7.604 1.436

aAugust 3, 2021 Exchange Rate: 1 US$ = 1,149 Won (South Korean currency).

(22)

13

Finally, the Stone price index approach developed by Deaton and Muellbauer (1980) was implemented for the construction of three prices for M1 (terrestrial, seafood, and plant protein) and four prices for M2 (fish, cephalopods, shellfish, and crustaceans). 𝑙𝑛 (𝑝_𝑖) is the Stone price index (Stone, 1954) of protein computed as:

𝑝_𝑘𝑡= 𝑝_𝑘,2015∙ 𝑐𝑝𝑖_𝑘𝑡, (13)

𝑙𝑛(𝑝_𝑖𝑡) = ∑_𝑘∈𝑖𝑤_𝑘𝑡𝑙𝑛(𝑝_𝑘𝑡), where 𝑖 =

{terrestrial meat, seafood, plant-based protein for M1 fish, cephalopods, molluscs, crustaceans for M2 .(14)

where 𝑝𝑘,2015 are the unit values from January 2015 for subgroup 𝑘, and 𝑐𝑝𝑖𝑘𝑡 and 𝑤𝑘𝑡 are the CPI and expenditure shares for subgroup 𝑘 and period 𝑡, respectively.

4.2. Incorporating sociodemographic factors

The demand for a particular good also depends on sociodemographic factors (Toufique et al., 2018). Accounting for sociodemographic factors when estimating a demand model is common with household data and can be found in several studies assessing seafood demand (e.g., Bronnmann et al., 2016; Buason &

Agnarsson, 2020; Dey et al., 2011; Salvanes & Devoretz, 1997;

Surathkal et al., 2017; Toufique et al., 2018). In this study, the presence of children, the household head’s age, and respondents’health-related factors, (namely BMI and HC), were included in the analyses.

BMI is an important indicator of health status (Braha et al., 2017) and has been widely used to indicate obesity (Kim et al., 2019;

(23)

14

Romero-Corral et al., 2008). Individuals with a BMI less than 18.5 are considered underweight⁶, while, those with a BMI in the range of 18.5–24.9 are of a normal weight. Those with a BMI in the range of 25.0–29.9 are considered overweight and otherwise obese (BMI≥ 30).

To assess HC, the scale of Kähkönen et al. (1996) was used. This scale has been used in several studies (e.g., Apaolaza et al., 2018;

Kähkönen & Tuorila, 1999; Sun, 2008). HC captures respondents’

concerns regarding health-related issues. In this study, it was determined by asking the respondents a set of eight questions, using a 5-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree).

Confirmatory factor analysis (CFA) is conducted on a set of HC items (Nunnally, 1994). It generates factor loadings of HC, which can be used as a single index. The Cronbach’s 𝛼 and composite reliability values for the HC construct for both models were robust and above the lower limit of 0.6, supporting the reliability for both models (internal consistency). Bartlett’s sphericity tests were significant for both models (p < 0.01), indicating that the data was sufficiently correlated. The Kaiser-Meyer-Olkin (KMO) values of M1 and M2 suggest adequate suitability for factor analysis (Cerny &

Kaiser, 1977). This study omitted HC5 and HC7, the factor loadings of which fell below 0.5, to maintain content validity but ensure convergent validity. The standardized factor loadings of the remaining items were significant (p < 0.01). The average variance extracted (AVE) of the HC constructs of M1 and M2 exceeded the minimum criterion of 0.5, indicating that the measures share more than 50% of their variation with the latent variable. Descriptions of the HC items and CFA results are provided in Table A1 in the Appendix.

6 Although these BMI categories are widely used, they may need to be adjusted upward in the near future to accommodate for population-based changes in height and weight (e.g., Nuttall, 2015).

(24)

15

In QUAIDS model, the sociodemographic variables of the households enter the demand system through the intercept.

Following Pollak and Wales (1981), this study modelled the intercept as linear combinations of a set of sociodemographic variables observed in the data. This translating approach allows for the level of demand to depend upon sociodemographic variables and preserves the conditional linearity of the model (Lecocq & Robin, 2015). The equation to be estimated with the sociodemographic variables incorporated is then given as:

𝑤_𝑖 = 𝛼_𝑖0+ ∑^𝑛_𝑗=1𝛾_𝑖𝑗ln(𝑝_𝑗) + 𝛽_𝑖ln { ^𝐸

𝑎(𝑝)} + ^𝜃^𝑖

𝑏(𝑝)ln { ^𝐸

𝑎(𝑝)}² + ∑ 𝛿ℎ 𝑖ℎ𝑑𝑒𝑚𝑜ℎ+ 𝜀_𝑖, (15)

where 𝑑𝑒𝑚𝑜ℎ is the ℎ^th sociodemographic variable, of which there are four: the presence of children, the age of the household head, BMI, and HC. Table A2 provides definitions of the sociodemographic variables and average monthly expenditure shares for each consumer segment of M1 and M2, respectively.

4.3. Locating seasonal cycles

The availability of seafood may act as a barrier to seafood protein consumption. If there is a lack of products available for the desired species, other available species can be a weak substitute for the preferred species (Torrissen & Onozaka, 2017). The availability of perishable agricultural products can be affected by seasonal production cycles (Arnade et al., 2004). Thus, the seasonal structure must be accounted for and isolated to improve the accuracy of demand analysis for the fresh seafood industry (Arnade et al., 2004).

Figures 1 and 2 respectively demonstrate the development of the monthly average expenditure shares and aggregated monthly prices

(25)

16

of the three protein sources. There are seasonality patterns in the expenditure shares and prices of the protein sources; these were found to differ upon comparison. Previous studies have shown the seasonality of seafood consumption and price, depicting an inverse relationship between the two (e.g., Dey et al., 2011). However, the relationship between seafood consumption and price is systematic, yet too complex to be viewed simply as an inverse relationship. To illustrate this point, the expenditure shares of seafood in the dataset generally peak every November (ranging from 24% to 26%), and hit bottoms every July (ranging from 17% to 20%) (Figure 1). On the other hand, aggregated prices of seafood show higher values in November (US$8.4/kg–US$9.0/kg), and September (US$6.9/kg–

US$8.2/kg) (Figure 2). Likewise, each seafood category in M2 exhibits different seasonality patterns, and it should be noted that their prices cannot simply explain their expenditure shares.

(26)

17

aAugust 3^rd, 2021 Exchange Rate: 1 US$ = 1,149 Won (South Korean currency)

Figure 1. Aggregated prices of the three protein sources over the data period.

Figure 2. Shares of the three protein sources in the total protein expenditures over the data period.

0 5 10 15 20

Jan Feb Mar Apr

May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr

May Jun Jul Aug Sep Oct Nov

2015 2016 2017 2018 2019

Price in $/kga

Terrestrial meat Seafood Plant-based protein

20%0%

40%60%

100%80%

Jan Feb Mar Apr

May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr

May Jun Jul Aug Sep Oct Nov

2015 2016 2017 2018 2019

Share

Terrestrial meat Seafood Plant-based protein

(27)

18

The monthly average expenditure shares and aggregated monthly prices of four seafood categories are depicted in Figures A1 and A2 in the Appendix.

When there are seasonal effects, and seasonal consumption is not explained solely by the price, the most common approach to account for seasonality has been the use of deterministic seasonal dummy variables (e.g., Bronnmann et al., 2016; Buason & Agnarsson, 2020; Tabarestani et al., 2017). However, previous studies (i.e., Arnade & Pick, 1998; Fraser & Moosa, 2002) have questioned the use of such an approach, pointing to the possibility of biased estimates. Since the introduction of different varieties and sources of supply and demand can affect the seasonal structure over time in unknown ways, it is crucial to control for the seasonal structure of demand models (Arnade et al., 2004).

To locate the seasonal cycle, the present study assumed that the demand follows a one-year cycle (Arnade et al., 2004). The intercept and demand equation for estimation with the seasonal trigonometric variables are then modified as:

𝑤_𝑖 = 𝛼_𝑖0+ ∑^𝑛_𝑗=1𝛾_𝑖𝑗ln(𝑝_𝑗) + 𝛽_𝑖ln {_𝑎(𝑝)^𝐸 } +_𝑏(𝑝)^𝜃^𝑖 ln {_𝑎(𝑝)^𝐸 }²

+ ∑ 𝛿_ℎ _𝑖ℎ𝑑𝑒𝑚𝑜_ℎ+ ∑ 𝜏_𝑡 _𝑖𝑠𝑠𝑒𝑎𝑠𝑜𝑛_𝑠+ 𝜀_𝑖, (16)

where 𝑠𝑒𝑎𝑠𝑜𝑛_𝑠 is the 𝑠^𝑡ℎ seasonality variable, of which there are three: 𝑠𝑒𝑎𝑠𝑜𝑛₁= 𝑡, 𝑠𝑒𝑎𝑠𝑜𝑛₂= 𝑐𝑜𝑠 ((2𝑓 𝑚⁄ )𝜋𝑡) , and 𝑠𝑒𝑎𝑠𝑜𝑛₃= 𝑠𝑖𝑛 ((2𝑓 𝑚⁄ )𝜋𝑡). Here, 𝑡 is the observation number. The value of 𝑓 corresponds to the seasonal frequencies of the data. In this paper, it is assumed that there is one cycle per year, 𝑓 = 1. The coefficients (𝜏_𝑖2, 𝜏_𝑖3) represent the contribution of each cycle to the seasonal process 𝑚. Since we use monthly data, 𝑚 = 12 (Arnade & Pick,

(28)

19 1998)⁷.

4.4. Handling endogeneity

The present study used the seemingly unrelated regression (SUR) procedure for its estimation. The SUR procedure takes the optimization process underlying the demand system through an adjustment for cross-equation contemporaneous correlation.

However, it generally did not provide consistent estimators for Equation (16) due to the potential endogeneity of some of the right- hand side variables. In each share equation, the error term, 𝜀_𝑖, can be correlated with the log of the total budget variable 𝑙𝑛𝐸. The correlation, a source of potential bias, can be accounted for with instrumental variables (IVs) (Hausman, 1978; Holly & Sargan, 1982).

In this study, it was assumed that the total expenditure is endogenous, leading to the creation of a logarithm of income as the identifying IV. Potential endogeneity in expenditure was accounted for by estimating a model for the total expenditure in each household and then incorporating the residuals of the model as an additional control variable. 𝜀_𝑖 in Equation (16) was augmented with the residual vector v̂ so that:

𝜀_𝑖 = 𝜌_𝑖v̂ + 𝑢_𝑖. (17)

The residual vector v̂ is gained from the first-stage IV regression.

Here, the dependent variable is the log of the total expenditure, and the independent variables are all exogenous variables entering the model and the identifying IV. From the first regression, residuals,

7 . For a more complete explanation of the trigonometric specification, see Arnade and Pick (1998), Arnade et al. (2004), and Canova and Hansen (1995).

(29)

20

namely, v̂, are included as a new variable in the final specification of the demand system:

𝑤_𝑖 = 𝛼_𝑖0+ ∑^𝑛_𝑗=1𝛾_𝑖𝑗ln(𝑝_𝑗) + 𝛽𝑖ln { ^𝐸

𝑎(𝑝)} +_𝑏(𝑝)^𝜃^𝑖 ln { ^𝐸

𝑎(𝑝)}² + ∑ 𝛿ℎ 𝑖ℎ𝑑𝑒𝑚𝑜ℎ+ ∑ 𝜏_𝑡 _𝑖𝑠𝑠𝑒𝑎𝑠𝑜𝑛_𝑠+ 𝜌_𝑖v̂ + 𝑢_𝑖, (18)

where 𝜌_𝑖 is the coefficient for v̂ of i^th goods, and 𝑢_𝑖 is an error term.

The descriptive statistics for the variables estimated and interpreted through the analyses are summarized in Table 2. In M1, the price of terrestrial protein ranges in the upper price segment, averaging about US$15/kg. As expected, the price of plant protein ranges in the lower price segment (US$3/kg). Among the seafood protein categories (M2), shellfish are in the upper price segment (US$7/kg), and cephalopods are in the lower price segment (US$4/kg). Most of the protein budget is spent on terrestrial protein with just over 20% of the total expenditure on protein is spent on seafood. Moreover, fish accounted for the highest expenditure share (42%), followed by shellfish (35%). Regarding the sociodemographic variables, about 70% of the households raise children. The average household head in the sample is 56 years old. The BMI and HC estimates indicate that the average household heads are of a normal weight and have moderate health concerns.

(30)

21

Table 2. Variables and their descriptive statistics.

M1 M2

Variable Mean

Std.

Dev. Mean

Std.

Dev.

Terrestrial protein

Price ($/kg) 15.192 8.644 - - Expenditure

share 0.669 0.305 - -

Seafood protein

share 0.203 0.247 - -

Fish Price ($/kg) - - 6.427 5.315 Expenditure

share - - 0.420 0.408

Cephalopods Price ($/kg) - - 3.768 5.234 Expenditure

share - - 0.192 0.319

Shellfish Price ($/kg) - - 6.892 11.814 Expenditure

share - - 0.249 0.353

Crustaceans Price ($/kg) - - 4.171 7.411 Expenditure

share - - 0.139 0.286

Plant protein

share 0.129 0.221 - -

Real

expenditure

(31)

22 Total protein

sources 64.157 54.964 - -

Total

Seafood - - 20.443 21.908

Sociodemo- graphic variables

Presence of

children^a 0.721 0.449 0.724 0.447 Age^b 55.956 9.824 56.025 9.789 BMI^c 22.321 2.629 22.333 2.618 HC^d 3.718 0.720 3.725 0.712

Number of observations 35,654 22,643

Number of households 634 631

a Dummy variable, 1= have kids.

b The variable is converted into an ordinal variable: coded as 1 if age ≤ 30; 2 if 30 < age ≤ 40; 3 if 40 < age ≤ 50; 4 if 50 < age ≤ 60; 5 if 60 < age ≤ 70; and 5 if age > 70.

c The variable is converted into an ordinal variable: coded as 1 if BMI <

18.5; 2 if 18.5 ≤ BMI < 25; 3 if 25 ≤ BMI < 30; and 4 if BMI ≥ 30.

d The variable is converted into a latent variable score. The table shows the average of the HC items.

Chapter 5. Empirical results

The goodness-of-fit statistics, and their associated p-values in the demand estimations for M1 and M2 are presented in Table 3. The explanatory power of the equations in M1 ranges from 56% to 65%

of the variance. The R² values of the equations in M2 are relatively higher than those of the equations in M1, ranging from 76% to 83%.

Thus, each equation in both M1 and M2 describes a considerable amount of variability in each dependent variable.

(32)

23

Table 3. Goodness-of-fit statistics of M1 and M2.

Model Equation RMSE R² p-value

M1

Terrestrial protein 0.173 0.680 0.000 Seafood protein 0.163 0.562 0.000 Plant protein 0.131 0.651 0.000

M2

Fish 0.168 0.831 0.000

Cephalopods 0.155 0.766 0.000

Shellfish 0.164 0.785 0.000

Crustaceans 0.135 0.778 0.000

The current study tested the exogeneity of expenditures in the whole demand system of M1 and M2 using the procedure developed by (Hausman, 1978). To apply a Hausman specification test, Wald tests for the parameters of each residual variable and the joint parameters were conducted. The Wald tests for M1 and M2 indicated that all expenditures are endogenous except the fish expenditure variables. These results are presented in Table 4.

(33)

24

Table 4. Hausman specification tests for M1 and M2.

Model Category Test χ² Pr > χ² Label

M1

Terrestrial protein Wald 252.799 0.000 𝜌₁= 0 Seafood protein Wald 90.909 0.000 𝜌₂= 0 Plant protein Wald 62.785 0.000 𝜌₃= 0 Joint test for two

categories (excludes plant protein)

Wald 260.410 0.000 𝜌1, 𝜌2= 0

M2

Fish Wald 0.342 0.559 𝜌1= 0

Cephalopods Wald 142.289 0.000 𝜌2= 0

Shellfish Wald 104.308 0.000 𝜌3= 0

Crustaceans Wald 629.235 0.000 𝜌4= 0 Joint test for three

categories (excludes cephalopods)

Wald 721.106 0.000 𝜌1, 𝜌3, 𝜌4

= 0

5.1. Demand for three major protein sources

The estimated parameters of M1 are presented in Table 5.

People who purchase more terrestrial meat than the average consumer are relatively younger and have children. The BMI parameter is positive, which indicates that these consumers are not as healthy as those who are more likely to consume other protein sources. Those consuming more seafood protein are typically older, healthier individuals with no children. Those who consume the most plant protein are generally older, healthier, and more health- conscious individuals with children.

(34)

25

Table 5. Estimation results of the QUAIDS from M1.

Terrestrial protein

Seafood protein

Plant protein

Intercept 𝛼𝑖0

5.123^***

(0.052)

1.896^***

(0.062)

–6.019^***

(0.018) ln Expenditure 𝛽_𝑖 0.046^***

(0.001)

0.017^***

(0.001)

–0.063^***

(0.000) ln Price of

terrestrial protein 𝛾_𝑖1 –0.011^***

(0.002)

–0.052^***

(0.002)

0.063^***

(0.003) ln Price of seafood

protein 𝛾𝑖2

–0.052^***

(0.003)

0.027^***

(0.001)

0.025^***

(0.004) ln Price of plant

protein 𝛾_𝑖3 0.063^***

(0.002)

0.025^***

(0.001)

–0.088^***

(0.001) Presence of

children 𝛿𝑖1

0.047^***

(0.002)

–0.031^***

(0.002)

–0.016^***

(0.002)

Age 𝛿_𝑖2 –0.024^***

(0.001)

0.017^***

(0.001)

0.008^***

(0.001)

BMI 𝛿_𝑖3 0.015^***

(0.002)

–0.009^***

(0.002)

–0.006^***

(0.002) Health Concern 𝛿_𝑖4 –0.000

(0.001)

–0.001 (0.001)

0.002^**

(0.001)

t 𝜏_1𝑖 –0.000

(0.000)

0.000^***

(0.000)

–0.000 (0.000)

cos^a 𝜏2𝑖

–0.007^***

(0.001)

0.006^***

(0.001)

0.002 (0.001)

sin^a 𝜏_3𝑖 0.007^***

(0.001)

–0.005^***

(0.001)

–0.002 (0.001) Residuals 𝜌_𝑖 0.021^***

(0.001)

–0.013^***

(0.001)

–0.008^***

(0.001)

Observations 35,654

Adj. R² 0.520

***p < 0.01, ^**p < 0.05, ^*p < 0.10. Standard errors are in parentheses.

a cos denotes 𝑠𝑒𝑎𝑠𝑜𝑛₂= cos ((𝑘 𝑍⁄ )πt) and sin denotes 𝑠𝑒𝑎𝑠𝑜𝑛₃= sin ((𝑘 𝑍⁄ )πt).

(35)

26

To address the economic interpretation of the estimated parameters in M1, this study derived expenditure, uncompensated, and compensated price elasticities. Table 6 shows the elasticity estimates. Statistically, all elasticities are significant.

Both animal proteins, terrestrial and seafood protein, demonstrate strongly elastic expenditure responsiveness and are luxuries. In particular, seafood protein shows more elastic expenditure responsiveness than terrestrial protein. Expenditure elasticities for the plant protein category imply that it is a necessity.

Thus, holding all other factors constant, seafood protein would attract higher expenditures than its protein substitutes if the expenditure on proteins increases. Considering the current expenditure share of seafood protein relative to other protein sources, its higher expenditure elasticity, and the prospects for increased seafood consumption (OECD-FAO, 2016), seafood protein has strong potential for further demand growth.

The uncompensated and compensated own-price elasticity estimates reveal a negative relationship between prices of normal goods and their demand. Comparing the uncompensated elasticities with the corresponding compensated values thus show the role of the two effects of price change on demand. All compensated own-price elasticities are smaller in absolute value than the corresponding uncompensated values. The results imply that the price responsiveness of the different protein sources is dependent on income; when income is held constant (i.e., it is not a constraint in the decision process), consumers tend to be less responsive to the price of the protein categories. The own-price elasticity of seafood protein is found to be relatively inelastic in comparison with the other two protein categories.

The compensated cross-price elasticities show that the plant protein categories are mutual substitutes with terrestrial and seafood protein. The degree of substitutability has increased compared with

(36)

27

the corresponding uncompensated elasticities. On the other hand, the terrestrial and seafood proteins are found to be mutual complements.

Table 6. Expenditure elasticities and price elasticities from M1.

Terrestrial protein

Seafood protein

Plant protein Expenditure elasticities 1.168^***

(0.005)

1.274^***

(0.034)

0.906^***

(0.000) Uncompensated (Marshallian) price elasticities

Terrestrial protein –1.900^***

(0.027)

–0.495^***

(0.012)

1.226^***

(0.035) Seafood protein –2.268^***

(0.267)

–1.040^***

(0.027)

2.034^***

(0.260)

Plant protein 0.576^***

(0.001)

0.207^***

(0.000)

–1.689^***

(0.001) Compensated (Hicksian) price elasticities

Terrestrial protein –1.580^***

(0.001)

–0.424^***

(0.006)

2.004^***

(0.037) Seafood protein –1.919^***

(0.252)

–0.963^***

(0.032)

2.882^***

(0

저작자표시

Master’s Thesis of Economics

Graduate School of

Graduate School of

Preface

Table of Contents

I. Essay 1: Modelling the demand across three major protein sources: focusing on

Chapter 2. Previous literature

Chapter 3. The QUAIDS framework

Chapter 4. Data and empirical specification

4.1. Demand system estimation in the absence of price data

4.2. Incorporating sociodemographic factors

4.3. Locating seasonal cycles

4.4. Handling endogeneity

Chapter 5. Empirical results

5.1. Demand for three major protein sources